Automated Vehicle Multi-Object Tracking at Scale with CAN

Abstract

Millions of vehicles are on the road with RADAR sensors in use for adaptive cruise control (ACC), and RADAR sen-sors are not tracking all of the objects in the field of view. This work shows a work-in-progress tool to improve tracking from RADAR and controller area network (CAN) which should be vitally useful for safety of transportation systems and automated vehicle development. The CAN data provides object detections, but there is a lingering data association problem. The contribution of this work in progress is the solution to the data association problem by posing the data association as a minimum cost network flow problem, and doing it at low cost with an eye toward scalable CPS research.